docker: periodically reconcile containers

drivers/docker/config.go

@@ -134,6 +134,25 @@ var (
 		Name:              pluginName,
 	}
 
+	danglingContainersBlock = hclspec.NewObject(map[string]*hclspec.Spec{
+		"enabled": hclspec.NewDefault(
+			hclspec.NewAttr("enabled", "bool", false),
+			hclspec.NewLiteral(`true`),
+		),
+		"period": hclspec.NewDefault(
+			hclspec.NewAttr("period", "string", false),
+			hclspec.NewLiteral(`"5m"`),
+		),
+		"creation_grace": hclspec.NewDefault(
+			hclspec.NewAttr("creation_grace", "string", false),
+			hclspec.NewLiteral(`"5m"`),
+		),
+		"dry_run": hclspec.NewDefault(
+			hclspec.NewAttr("dry_run", "bool", false),
+			hclspec.NewLiteral(`false`),
+		),
+	})
+
 	// configSpec is the hcl specification returned by the ConfigSchema RPC
 	// and is used to parse the contents of the 'plugin "docker" {...}' block.
 	// Example:
@@ -195,6 +214,10 @@ var (
 				hclspec.NewAttr("container", "bool", false),
 				hclspec.NewLiteral("true"),
 			),
+			"dangling_containers": hclspec.NewDefault(
+				hclspec.NewBlock("dangling_containers", false, danglingContainersBlock),
+				hclspec.NewLiteral("{}"),
+			),
 		})), hclspec.NewLiteral(`{
 			image = true
 			container = true
@@ -491,6 +514,28 @@ type DockerVolumeDriverConfig struct {
 	Options hclutils.MapStrStr `codec:"options"`
 }
 
+// ContainerGCConfig controls the behavior of the GC reconciler to detects
+// dangling nomad containers that aren't tracked due to docker/nomad bugs
+type ContainerGCConfig struct {
+	// Enabled controls whether container reconciler is enabled
+	Enabled bool `codec:"enabled"`
+
+	// DryRun indicates that reconciler should log unexpectedly running containers
+	// if found without actually killing them
+	DryRun bool `codec:"dry_run"`
+
+	// PeriodStr controls the frequency of scanning containers
+	PeriodStr string        `codec:"period"`
+	period    time.Duration `codec:"-"`
+
+	// CreationGraceStr is the duration allowed for a newly created container
+	// to live without being registered as a running task in nomad.
+	// A container is treated as leaked if it lived more than grace duration
+	// and haven't been registered in tasks.
+	CreationGraceStr string        `codec:"creation_grace"`
+	CreationGrace    time.Duration `codec:"-"`
+}
+
 type DriverConfig struct {
 	Endpoint        string       `codec:"endpoint"`
 	Auth            AuthConfig   `codec:"auth"`
@@ -519,6 +564,8 @@ type GCConfig struct {
 	ImageDelay         string        `codec:"image_delay"`
 	imageDelayDuration time.Duration `codec:"-"`
 	Container          bool          `codec:"container"`
+
+	DanglingContainers ContainerGCConfig `codec:"dangling_containers"`
 }
 
 type VolumeConfig struct {
@@ -534,6 +581,8 @@ func (d *Driver) ConfigSchema() (*hclspec.Spec, error) {
 	return configSpec, nil
 }
 
+const danglingContainersCreationGraceMinimum = 1 * time.Minute
+
 func (d *Driver) SetConfig(c *base.Config) error {
 	var config DriverConfig
 	if len(c.PluginConfig) != 0 {
@@ -551,6 +600,25 @@ func (d *Driver) SetConfig(c *base.Config) error {
 		d.config.GC.imageDelayDuration = dur
 	}
 
+	if len(d.config.GC.DanglingContainers.PeriodStr) > 0 {
+		dur, err := time.ParseDuration(d.config.GC.DanglingContainers.PeriodStr)
+		if err != nil {
+			return fmt.Errorf("failed to parse 'period' duration: %v", err)
+		}
+		d.config.GC.DanglingContainers.period = dur
+	}
+
+	if len(d.config.GC.DanglingContainers.CreationGraceStr) > 0 {
+		dur, err := time.ParseDuration(d.config.GC.DanglingContainers.CreationGraceStr)
+		if err != nil {
+			return fmt.Errorf("failed to parse 'creation_grace' duration: %v", err)
+		}
+		if dur < danglingContainersCreationGraceMinimum {
+			return fmt.Errorf("creation_grace is less than minimum, %v", danglingContainersCreationGraceMinimum)
+		}
+		d.config.GC.DanglingContainers.CreationGrace = dur
+	}
+
 	if c.AgentConfig != nil {
 		d.clientConfig = c.AgentConfig.Driver
 	}
@@ -568,6 +636,9 @@ func (d *Driver) SetConfig(c *base.Config) error {
 
 	d.coordinator = newDockerCoordinator(coordinatorConfig)
 
+	reconciler := newReconciler(d)
+	reconciler.Start()
+
 	return nil
 }
 

drivers/docker/driver.go

@@ -66,6 +66,13 @@ var (
 	nvidiaVisibleDevices = "NVIDIA_VISIBLE_DEVICES"
 )
 
+const (
+	dockerLabelTaskID   = "com.hashicorp.nomad.task_id"
+	dockerLabelTaskName = "com.hashicorp.nomad.task_name"
+	dockerLabelAllocID  = "com.hashicorp.nomad.alloc_id"
+	dockerLabelJobName  = "com.hashicorp.nomad.job_name"
+)
+
 type Driver struct {
 	// eventer is used to handle multiplexing of TaskEvents calls such that an
 	// event can be broadcast to all callers
@@ -309,6 +316,10 @@ CREATE:
 		// the container is started
 		runningContainer, err := client.InspectContainer(container.ID)
 		if err != nil {
+			client.RemoveContainer(docker.RemoveContainerOptions{
+				ID:    container.ID,
+				Force: true,
+			})
 			msg := "failed to inspect started container"
 			d.logger.Error(msg, "error", err)
 			client.RemoveContainer(docker.RemoveContainerOptions{
@@ -977,9 +988,19 @@ func (d *Driver) createContainerConfig(task *drivers.TaskConfig, driverConfig *T
 
 	if len(driverConfig.Labels) > 0 {
 		config.Labels = driverConfig.Labels
-		logger.Debug("applied labels on the container", "labels", config.Labels)
 	}
 
+	config.Labels = map[string]string{
+		dockerLabelTaskID:   task.ID,
+		dockerLabelTaskName: task.Name,
+		dockerLabelAllocID:  task.AllocID,
+		dockerLabelJobName:  task.JobName,
+	}
+	for k, v := range driverConfig.Labels {
+		config.Labels[k] = v
+	}
+	logger.Debug("applied labels on the container", "labels", config.Labels)
+
 	config.Env = task.EnvList()
 
 	containerName := fmt.Sprintf("%s-%s", strings.Replace(task.Name, "/", "_", -1), task.AllocID)

drivers/docker/driver_test.go

@@ -905,7 +905,8 @@ func TestDockerDriver_Labels(t *testing.T) {
 		t.Fatalf("err: %v", err)
 	}
 
-	require.Equal(t, 2, len(container.Config.Labels))
+	// expect to see 4 additional standard labels
+	require.Equal(t, len(cfg.Labels)+4, len(container.Config.Labels))
 	for k, v := range cfg.Labels {
 		require.Equal(t, v, container.Config.Labels[k])
 	}
@@ -1008,6 +1009,38 @@ func TestDockerDriver_CreateContainerConfig(t *testing.T) {
 	require.Equal(t, containerName, c.Name)
 }
 
+func TestDockerDriver_CreateContainerConfig_Labels(t *testing.T) {
+	t.Parallel()
+
+	task, cfg, _ := dockerTask(t)
+	task.AllocID = uuid.Generate()
+	task.JobName = "redis-demo-job"
+
+	cfg.Labels = map[string]string{
+		"user_label": "user_value",
+	}
+
+	require.NoError(t, task.EncodeConcreteDriverConfig(cfg))
+
+	dh := dockerDriverHarness(t, nil)
+	driver := dh.Impl().(*Driver)
+
+	c, err := driver.createContainerConfig(task, cfg, "org/repo:0.1")
+	require.NoError(t, err)
+
+	expectedLabels := map[string]string{
+		// user provided labels
+		"user_label": "user_value",
+		// default labels
+		"com.hashicorp.nomad.alloc_id":  task.AllocID,
+		"com.hashicorp.nomad.job_name":  "redis-demo-job",
+		"com.hashicorp.nomad.task_id":   task.ID,
+		"com.hashicorp.nomad.task_name": "redis-demo",
+	}
+
+	require.Equal(t, expectedLabels, c.Config.Labels)
+}
+
 func TestDockerDriver_CreateContainerConfig_Logging(t *testing.T) {
 	t.Parallel()